Go to Qiita Advent Calendar 2024 Top
LoginSignup
0
0

Delete article

Deleted articles cannot be recovered.

Draft of this article would be also deleted.

Are you sure you want to delete this article?

More than 1 year has passed since last update.

@Maki-HamarukiLab(Maki Sunwood AI Labs.)

データサイエンスでDota2強くなるかも説(5)～ハード試合データを解析してみたら驚愕の事実が～

Posted at

はじめに

最近，Dota2を始めましたが全く勝てません
ハードボットにボコボコにされます．

色々と調べても「死ぬな」くらいのことしか分らず苦戦しています．

データサイエンスでDota2強くなるかも説

そこで，データサイエンスの力を借りて，どのような状況なら勝っているか？や前回に比べてどのように振舞ったから勝てたのか？ということを数値化して分析していけば強くなるのでは！と考えました．本企画はその仮説を検証していく企画です．

file

前回までのあらすじ

環境構築編

Dota2の情報をPythonで取得できるような環境を作成し*1，データを取得＋定期的に保存する機構を作りました*2．

データ解析編

実際にBotとの対戦をプロットしてみて客観的に解析してみたところ*3，ミディアムボットとの対戦をしてもハードボットとの対戦ではそこまで効果がない可能性が発見されました*4．

今回の概要

ハード試合データを解析してみたいと思います．

必要なもの

Dota2

Dota2解析環境

Dota2 解析プログラム

使用データセット

  • bot戦（ハード）：Jakiro【勝利】
  • bot戦（ハード）：Jakiro【勝利】
  • bot戦（ハード）：Jakiro【敗北】
  • bot戦（ハード）：Jakiro【勝利】
  • bot戦（ハード）：Jakiro【勝利】
  • bot戦（ハード）：Jakiro【勝利】
  • bot戦（ハード）：Jakiro【敗北】
  • bot戦（ハード）：Jakiro【敗北】

8試合のデータセットを使います．いずれもjakiroを使用しました．
ハードボットとの対戦について，僕的にはこいつが使いやすいかなと初心者ながら感じでいます．

Import package

import dota2gsi
import glob
import pandas as pd

import seaborn as sns
from matplotlib import pyplot as plt
import matplotlib as mpl

figure style


#sns.palplot(sns.color_palette("RdBu_r", 24))
#sns.set_palette("RdBu_r")
#sns.set(style='darkgrid')
#sns.set_style('whitegrid')
mpl.style.use('fivethirtyeight')

Read csv file

ログのフォルダを指定します．

target_dir = "logs"

ログのフォルダ内のcsvファイルを検索します．

#target_log_file_list = ["log_20220901-160140_BM_W.csv", "log_20220901-204148_BM_W.csv", "log_20220902-010822_BH_L.csv", "log_20220904-035245_BH_L.csv"]
target_log_file_list = glob.glob(target_dir + "/*.csv")

ファイルの一覧はこちらになります．

target_log_file_list

    ['logs\\log_20220904-161354_npc_dota_hero_jakiro_W.csv',
     'logs\\log_20220905-092008_npc_dota_hero_jakiro_W.csv',
     'logs\\log_20220905-131433_npc_dota_hero_jakiro_L.csv',
     'logs\\log_20220905-141336_npc_dota_hero_jakiro_W.csv',
     'logs\\log_20220905-151758_npc_dota_hero_jakiro_W.csv',
     'logs\\log_20220905-171404_npc_dota_hero_jakiro_W.csv',
     'logs\\log_20220906-093622_npc_dota_hero_jakiro_L.csv',
     'logs\\log_20220906-102435_npc_dota_hero_jakiro_L.csv']

ファイルを順に読みだしてDataFrame型をlistに追加していきます．

df_list = []
for file_path in target_log_file_list:
    df_list.append(pd.read_csv(file_path))

中身のDataFrameはこんな感じになります．

df_list[0].head(5)
Unnamed: 0 draft_activeteam draft_activeteam_time_remaining draft_dire_bonus_time draft_pick draft_radiant_bonus_time draft_team#:home_team hero_alive hero_break hero_buyback_cost ... provider_appid provider_version provider_timestamp hero_level_try0 hero_health_try0 player_gold_try0 player_camps_stacked_try0 hero_level_try0_W hero_level_W player_gold_try
0 0 NaN NaN NaN NaN NaN NaN True False 253 ... 570 47 1662275633 1 740 5.0 NaN 1 1 5.0
1 0 NaN NaN NaN NaN NaN NaN True False 253 ... 570 47 1662275635 1 740 5.0 NaN 1 1 5.0
2 0 NaN NaN NaN NaN NaN NaN True False 253 ... 570 47 1662275635 1 740 5.0 NaN 1 1 5.0
3 0 NaN NaN NaN NaN NaN NaN True False 253 ... 570 47 1662275635 1 740 5.0 NaN 1 1 5.0
4 0 NaN NaN NaN NaN NaN NaN True False 253 ... 570 47 1662275635 1 740 5.0 NaN 1 1 5.0

5 rows × 84 columns

DataFrameの列名の一覧はこちら

df_list[0].columns

    Index(['Unnamed: 0', 'draft_activeteam', 'draft_activeteam_time_remaining',
           'draft_dire_bonus_time', 'draft_pick', 'draft_radiant_bonus_time',
           'draft_team#:home_team', 'hero_alive', 'hero_break',
           'hero_buyback_cost', 'hero_buyback_cooldown', 'hero_disarmed',
           'hero_has_debuff', 'hero_health', 'hero_health_percent', 'hero_hexed',
           'hero_id', 'hero_level', 'hero_magicimmune', 'hero_mana',
           'hero_mana_percent', 'hero_max_health', 'hero_max_mana', 'hero_muted',
           'hero_name', 'hero_respawn_seconds', 'hero_selected_unit',
           'hero_silenced', 'hero_stunned', 'hero_talent_1', 'hero_talent_2',
           'hero_talent_3', 'hero_talent_4', 'hero_talent_5', 'hero_talent_6',
           'hero_talent_7', 'hero_talent_8', 'hero_xpos', 'hero_ypos',
           'map_clock_time', 'map_daytime', 'map_dire_ward_purchase_cooldown',
           'map_game_state', 'map_game_time', 'map_name', 'map_matchid',
           'map_radiant_ward_purchase_cooldown', 'map_nightstalker_night',
           'map_roshan_state', 'map_roshan_state_end_seconds', 'map_win_team',
           'map_customgamename', 'player_assists', 'player_camps_stacked',
           'player_deaths', 'player_denies', 'player_gold', 'player_gold_reliable',
           'player_gold_unreliable', 'player_gpm', 'player_hero_damage',
           'player_kill_list:victimid_#', 'player_kill_streak', 'player_kills',
           'player_last_hits', 'player_net_worth', 'player_pro_name',
           'player_runes_activated', 'player_support_gold_spent',
           'player_wards_destroyed', 'player_wards_placed',
           'player_wards_purchased', 'player_xpm', 'provider_name',
           'provider_appid', 'provider_version', 'provider_timestamp'],
          dtype='object')

Plot level section

Extract level data

ここから対象の列だけ抽出してDataFrameに新たにぶち込んでいきます．

df_ex_data_list = []
for i, df in enumerate(df_list):
    target_name = 'hero_level'
    result_flag = target_log_file_list[i].split(".csv")[0].split("_")[-1]
    
    plot_name = '{}_try{}_{}'.format(target_name, i, result_flag)
    #plot_name = '{}_{}'.format(target_name, result_flag)
    
    df[plot_name] = df[target_name]
    df_ex_data_list.append(df[plot_name])

df_ex_merge = pd.concat(df_ex_data_list, axis=1)

レベルの列だけを抽出したDataFrameがこちらになります．

df_ex_merge
hero_level_try0_W hero_level_try1_W hero_level_try2_L hero_level_try3_W hero_level_try4_W hero_level_try5_W hero_level_try6_L hero_level_try7_L
0 1.0 NaN NaN NaN NaN NaN NaN 1.0
1 1.0 NaN NaN NaN NaN NaN NaN 1.0
2 1.0 NaN NaN NaN NaN NaN NaN 1.0
3 1.0 NaN NaN NaN NaN NaN NaN 1.0
4 1.0 NaN NaN NaN NaN NaN NaN 1.0
... ... ... ... ... ... ... ... ...
33798 NaN NaN NaN NaN 23.0 NaN NaN NaN
33799 NaN NaN NaN NaN 23.0 NaN NaN NaN
33800 NaN NaN NaN NaN 23.0 NaN NaN NaN
33801 NaN NaN NaN NaN 23.0 NaN NaN NaN
33802 NaN NaN NaN NaN 23.0 NaN NaN NaN

33803 rows × 8 columns

Plot level data

レベルの遷移をプロットしたグラフがこちらです．

plt.figure(figsize=(20,8))
sns.lineplot(data=df_ex_merge, lw=2)
#sns.lineplot(data=df_data, x='clock_time', y='level', lw=2)

Mean data

make mean data

「勝ち」，「負け」のグループに分けて平均を算出していきます．

def extract_columns_data(df_ex_merge, target_name):
    columns_list = [i for i in df_ex_merge.columns if(target_name in i)]
    df_ex_merge[target_name] = df_ex_merge[columns_list].mean(axis=1, skipna=True)
    #df_ex_merge[target_name] = df_ex_merge.mean(axis=1)
    return df_ex_merge

df_mean_merge = extract_columns_data(df_ex_merge, "_W")
df_mean_merge = extract_columns_data(df_mean_merge, "_L")
df_mean_merge
player_gold_try0 player_gold_try1 player_gold_try2 player_gold_try3 player_gold_try4 player_gold_try5 player_gold_try6 player_gold_try7 _W _L
0 5.0 600.0 NaN NaN 5.0 NaN NaN 5.0 NaN NaN
1 5.0 600.0 600.0 600.0 5.0 NaN 600.0 5.0 NaN NaN
2 5.0 600.0 600.0 600.0 5.0 600.0 600.0 5.0 NaN NaN
3 5.0 600.0 600.0 600.0 5.0 600.0 600.0 5.0 NaN NaN
4 5.0 600.0 600.0 600.0 5.0 600.0 600.0 5.0 NaN NaN
... ... ... ... ... ... ... ... ... ... ...
33798 NaN NaN NaN NaN 4652.0 NaN NaN NaN NaN NaN
33799 NaN NaN NaN NaN 4652.0 NaN NaN NaN NaN NaN
33800 NaN NaN NaN NaN 4652.0 NaN NaN NaN NaN NaN
33801 NaN NaN NaN NaN 4652.0 NaN NaN NaN NaN NaN
33802 NaN NaN NaN NaN 4685.0 NaN NaN NaN NaN NaN

33803 rows × 10 columns

「勝ち」，「負け」を抽出して平均を算出したDataFrameがこちらです．

df_mean_merge2 = df_mean_merge[["_W", "_L"]]
df_mean_merge2
_W _L
0 1.0 1.0
1 1.0 1.0
2 1.0 1.0
3 1.0 1.0
4 1.0 1.0
... ... ...
33798 23.0 NaN
33799 23.0 NaN
33800 23.0 NaN
33801 23.0 NaN
33802 23.0 NaN

33803 rows × 2 columns

各試合のレベル遷移と勝ったときの平均，負けたときの平均をプロットしたデータがこちらです．

plt.figure(figsize=(20,12))


ax = sns.lineplot(data=df_mean_merge2, lw=4)
ax = sns.lineplot(data=df_ex_merge, lw=2, ax=ax)
plt.legend(labels=["Win","Lose"])
#sns.lineplot(data=df_data, x='clock_time', y='level', lw=2)

Plot health section

Extract health data

ここから体力の列だけ抽出してDataFrameに新たにぶち込んでいきます．

df_ex_data_list = []
for i, df in enumerate(df_list):
    target_name = 'hero_health'
    plot_name = '{}_try{}'.format(target_name, i)
    df[plot_name] = df[target_name]
    df_ex_data_list.append(df[plot_name])

df_ex_merge = pd.concat(df_ex_data_list, axis=1)

Plot health data

plt.figure(figsize=(20,5))
sns.lineplot(data=df_ex_merge, lw=2)

Plot gold data

Extract gold data

ここからゴールドの列だけ抽出してDataFrameに新たにぶち込んでいきます．

df_ex_data_list = []
for i, df in enumerate(df_list):
    target_name = 'player_gold'
    result_flag = target_log_file_list[i].split(".csv")[0].split("_")[-1]
    
    plot_name = '{}_try{}_r_{}'.format(target_name, i, result_flag)
    #plot_name = '{}_try'.format(target_name, i)
    
    df[plot_name] = df[target_name]
    df_ex_data_list.append(df[plot_name])

df_ex_merge = pd.concat(df_ex_data_list, axis=1)

ゴールドの遷移から，稼いだ金額の積算を算出します．

df_ex_merge2 = df_ex_merge.diff().fillna(0)
df_ex_merge2[df_ex_merge2<0] = 0
df_ex_merge2 = df_ex_merge2.cumsum()
df_ex_merge2

積算の結果がこちら

player_gold_try0_r_W player_gold_try1_r_W player_gold_try2_r_L player_gold_try3_r_W player_gold_try4_r_W player_gold_try5_r_W player_gold_try6_r_L player_gold_try7_r_L
0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
1 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
2 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
3 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
4 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
... ... ... ... ... ... ... ... ...
33798 17763.0 14212.0 14526.0 14759.0 18162.0 14715.0 21314.0 19372.0
33799 17763.0 14212.0 14526.0 14759.0 18162.0 14715.0 21314.0 19372.0
33800 17763.0 14212.0 14526.0 14759.0 18162.0 14715.0 21314.0 19372.0
33801 17763.0 14212.0 14526.0 14759.0 18162.0 14715.0 21314.0 19372.0
33802 17763.0 14212.0 14526.0 14759.0 18195.0 14715.0 21314.0 19372.0

33803 rows × 8 columns

勝，負の列をそれぞれ平均します．

df_mean_merge = extract_columns_data(df_ex_merge2, "r_W")
df_mean_merge = extract_columns_data(df_mean_merge, "r_L")
df_mean_merge
player_gold_try0_r_W player_gold_try1_r_W player_gold_try2_r_L player_gold_try3_r_W player_gold_try4_r_W player_gold_try5_r_W player_gold_try6_r_L player_gold_try7_r_L r_W r_L
0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
1 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
2 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
3 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
4 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
... ... ... ... ... ... ... ... ... ... ...
33798 17763.0 14212.0 14526.0 14759.0 18162.0 14715.0 21314.0 19372.0 15922.2 18404.0
33799 17763.0 14212.0 14526.0 14759.0 18162.0 14715.0 21314.0 19372.0 15922.2 18404.0
33800 17763.0 14212.0 14526.0 14759.0 18162.0 14715.0 21314.0 19372.0 15922.2 18404.0
33801 17763.0 14212.0 14526.0 14759.0 18162.0 14715.0 21314.0 19372.0 15922.2 18404.0
33802 17763.0 14212.0 14526.0 14759.0 18195.0 14715.0 21314.0 19372.0 15928.8 18404.0

33803 rows × 10 columns

勝，負の列だけを抽出します．

df_mean_merge2 = df_mean_merge[["r_W", "r_L"]]
df_mean_merge2
r_W r_L
0 0.0 0.0
1 0.0 0.0
2 0.0 0.0
3 0.0 0.0
4 0.0 0.0
... ... ...
33798 15922.2 18404.0
33799 15922.2 18404.0
33800 15922.2 18404.0
33801 15922.2 18404.0
33802 15928.8 18404.0

33803 rows × 2 columns

x軸にとるindexを作成します．

df_mean_merge3 = df_mean_merge2.reset_index()
df_mean_merge2
r_W r_L
0 0.0 0.0
1 0.0 0.0
2 0.0 0.0
3 0.0 0.0
4 0.0 0.0
... ... ...
33798 15922.2 18404.0
33799 15922.2 18404.0
33800 15922.2 18404.0
33801 15922.2 18404.0
33802 15928.8 18404.0

33803 rows × 2 columns

df_mean_merge2.columns

    Index(['r_W', 'r_L'], dtype='object')

Plot gold data

ゴールドの積算をプロットしたものです．

plt.figure(figsize=(20,10))
sns.lineplot(data=df_ex_merge2, lw=2)

平均もプロットしたものがこちらです．

plt.figure(figsize=(20,20))
#sns.lineplot(data=df_mean_merge2, x="index", y="player_gold_try")
ax = sns.lineplot(data=df_mean_merge2, lw=4)
sns.lineplot(data=df_ex_merge2, lw=2, ax=ax)
#plt.legend(labels=["Win","Lose"])

考察

8試合のデータを勝，負に分類して考察していこうと思います．

レベルとゴールドの遷移をプロットしたところ，いずれも勝利したときの平均のグラフの方が敗北したときの平均のグラフより下回っていました．これより，レベルやゴールドを順調に稼いだからといって勝てる訳ではないと考えられます．

これは，味方の振る舞いの影響が想定よりも大きく，自分のレベルやゴールドの稼ぎがそこまで影響していないということになります．
（とても悲しい．．．．）

プログラム

コードはこちらです．

おわりに

今回はハードボットとの対戦ログを解析してみた結果．
勝利している方が敗北している試合よりもレベル，ゴールドを多く稼いでいると思っていたのですが，その仮説とは相反した結果が得られました．敗北した方が勝利している方より稼いでいる結果になってました．

これは，味方の振る舞いの影響が想定よりも大きく，自分のレベルやゴールドの稼ぎがそこまで影響していないということになりました．

とても悲しい結果ですね．．．．

参考サイト

最新情報

速報はTwitterやブログにて

0
0
0

Register as a new user and use Qiita more conveniently

  1. You get articles that match your needs
  2. You can efficiently read back useful information
  3. You can use dark theme
What you can do with signing up
Sign upLogin
0
0

Delete article

Deleted articles cannot be recovered.

Draft of this article would be also deleted.

Are you sure you want to delete this article?